Inspection systems for the analysis of moving web materials have proven critical to modem manufacturing operations. Industries as varied as metal fabrication, paper, non-wovens, and films rely on these inspection systems for both product certification and online process monitoring. One major difficulty in the industry is related to the extremely high data processing rates required to keep up with current manufacturing processes. With webs of commercially viable width and web speeds that are typically used and pixel sizes that are typically needed, data acquisition speeds of tens or even hundreds of megabytes per second are required of the inspection systems. It is a continual challenge to process images and perform accurate defect detection at these data rates.
The art has responded to this dilemma by limiting the image processing to very simple algorithms, by limiting the scope and complexity of the detection algorithms, and by using custom inspection system architectures incorporating custom electronic hardware or dedicated preprocessors, each working on part of the data stream. While such systems are capable of achieving the data rates required for the inspection of moving webs, is very difficult to adapt the system for a new production process and web materials. Also, processing algorithms are limited to the capabilities of dedicated processing modules. Finally, as the image processing algorithms become more complex, the hardware required to implement the required processing quickly becomes unmanageable.
The manufacturing industry has recognized the importance of being able to produce product “just-in-time” with obvious advantages in reduced inventory. However, achieving this goal often has manufacturers working to develop systems and devices that allow a rapid change over between various products. The rapid changeover between products is inconsistent with the specialized signal processing hardware the art of optical inspection of moving webs now requires.
Another dilemma occurs in situations when a given product can be later used for multiple applications, with each of the multiple applications requiring different quality levels. The difficulty is that during the time of manufacture, it is not known which quality level will be required. Therefore, the current art attempts to grade quality level after defect detection by using various defect classification techniques based on spatial features of the extracted defects. While this is sometimes adequate when gross differences exist between defect levels for different quality requirements, it is not adequate for more demanding situations in which more subtle differences between defects require different image processing and defect extraction algorithms. Thus, if one waits until after defect extraction for classification, information is lost and the classification is impossible.